Fix UTF-8 offsets in sequence patterns

[yangfan-yf-yf]

Description

I reproduced a crash on master (f005ac786da7d556fdf5733e9b9ddae1eaf27242) when patching a sequence pattern containing non-ASCII text:

from rope.refactor import patchedast

source = "match value:\n    case [é, y]:\n        pass\n"
patchedast.get_patched_ast(source, True)

The walker reads : as the closing delimiter and raises MismatchedTokenError. Nested and empty sequence patterns after non-ASCII text are also affected, and some inputs produce incorrect source regions without raising.

AST columns are UTF-8 byte offsets, while the walker slices a Python string using character offsets. I convert the AST coordinates at the sequence-pattern slicing sites, including the empty-pattern path. The generic SourceLinesAdapter coordinate behavior is preserved.

The regression test covers seven patterns, including non-ASCII captures and literals, nested and empty sequences, bare sequences, and multiline input. It checks both source round-tripping and each sequence node's source region against ast.get_source_segment.

Related to #623; follows the sequence delimiter handling introduced in #863. Delimiter warnings for adjacent comments remain outside this change.

Validation on Windows / Python 3.12.14:

• All seven new regression scenarios fail before the fix and pass after it.
• Related modules: 311 passed, 10 skipped; seven subtests passed.
• Full suite: 2117 passed, 12 skipped, 5 xfailed; seven subtests passed.
• Target-file pre-commit checks and git diff --check pass.

Checklist

• I have added tests that prove my fix is effective.
• I have updated CHANGELOG.md.